Minicorso su Emerging Network Technologies
Ascolta
Si informano gli studenti interessati che la prof.ssa Tinnirello ha organizzato, assieme ai proff. Mangione e Croce, un minicorso su Emerging Network Technologies, le cui lezioni saranno tenute dal prof. Giovanni Neglia, nei giorni venerdi' 18, lunedi' 21 e martedi' 22 a partire dalle 15 in aula O011, sugli argomenti riportati in calce (per un totale di 8 ore).
Il minicorso verterà su argomenti un po' avanzati su alcune tecniche emergenti di machine learning, il cui programma è riportato in calce.
Verranno attribuiti 2 CFU agli studenti del Corso di Laurea Magistrale in Electronics Engineering che avranno seguito i seminari e superato la prova finale prevista.
-----
Programma:
- Introduction to federated learning (FL): motivations, cross-device, and cross-silo settings, differences with distributed training in clustering.
- Refresher about ML optimization (empirical risk vs expected risk, statistical learning guarantees).
- A look at the classic federated averaging algorithm (FedAvg). An overview of existing software frameworks and benchmarks for FL.
- Optimization Algorithms for FL: convergence results under iid and non-iid data.
- Model personalization to address statistical heterogeneity (different local datadistributions) and system heterogeneity (different clients’ capabilities).
- Communication: topology selection, sampling schemes, and compression.
- A look at the classic federated averaging algorithm (FedAvg). An overview of existing software frameworks and benchmarks for FL.
- Optimization Algorithms for FL: convergence results under iid and non-iid data.
- Model personalization to address statistical heterogeneity (different local datadistributions) and system heterogeneity (different clients’ capabilities).
- Communication: topology selection, sampling schemes, and compression.
- Privacy issues in FL